def set_drosophilia(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/drosophilia'
self.data_name = 'drosophilia'
self.results_dir = 'drosophilia'
self.data_set_file_name = 'split_data.pkl'
def set_boston_housing(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/boston_housing'
self.data_name = 'boston_housing'
self.results_dir = 'boston_housing'
self.data_set_file_name = 'split_data.pkl'
def set_adience_aligned_cnn_1(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/adience_aligned_cnn_1_per_instance_id'
self.data_name = 'adience_aligned_cnn_1_per_instance_id'
self.results_dir = 'adience_aligned_cnn_1_per_instance_id'
self.data_set_file_name = 'split_data.pkl'
def __init__(self, data_set=None):
super(ProjectConfigs, self).__init__()
self._num_labels = None
self.data_set = data_set
self.project_dir = 'base'
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = ''
self.data_name = ''
self.data_set_file_name = ''
self.results_dir = ''
self.include_name_in_results = False
self.labels_to_use = None
self.labels_to_not_sample = None
self.target_labels = None
self.source_labels = None
self.oracle_labels = None
self.num_labels = range(40, 201, 40)
#self.num_labels = range(40,81,40)
self.set_boston_housing()
self.num_splits = 30
self.labels_to_keep = None
self.labels_to_not_sample = {}
self.data_set = None
self.use_pool = False
self.pool_size = 2
self.method_results_class = results_lib.MethodResults
self.oracle_data_set_ids = None
def set_synthetic_linear_reg(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/synthetic_linear_reg500-50-1'
self.data_name = 'synthetic_linear_reg500-50-1'
self.results_dir = 'synthetic_linear_reg500-50-1'
self.data_set_file_name = 'split_data.pkl'
def set_data_path_results(self, name):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/' + name
self.data_name = name
self.results_dir = name
self.data_set_file_name = 'split_data.pkl'
def set_concrete(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/concrete'
self.data_name = 'concrete'
self.results_dir = 'concrete'
self.data_set_file_name = 'split_data.pkl'
def set_wine_red(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
s = 'wine-red'
self.data_dir = 'data_sets/' + s
self.data_name = s
self.results_dir = s
self.data_set_file_name = 'split_data.pkl'
def set_synthetic_regression(self, name):
self.loss_function = loss_function.MeanSquaredError()
self.target_labels = np.zeros(1)
self.source_labels = np.ones(1)
self.loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/' + name
self.data_name = name
self.results_dir = name
self.data_set_file_name = 'split_data.pkl'
def set_bike_sharing(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
assert self.use_1d_data == True
self.data_dir = 'data_sets/bike_sharing-feat=1'
self.data_name = 'bike_sharing-feat=1'
self.results_dir = 'bike_sharing-feat=1'
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([1])
self.source_labels = np.asarray([0])
def set_pollution(self, id, size):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
#assert self.use_1d_data == True
s = 'pollution-%d-%d-norm' % (id, size)
self.data_dir = 'data_sets/' + s
self.data_name = s
self.results_dir = s
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([0])
self.source_labels = np.asarray([1])
def __init__(self, configs=MethodConfigs()):
super(KDE, self).__init__(configs)
self.cv_params = {'sigma': np.asarray(10.0**np.asarray(range(-4, 5)))}
self.cv_params = {}
self.is_classifier = False
self._estimated_error = None
self.quiet = True
self.best_params = None
self.model = None
self.configs.loss_function = loss_function.MeanSquaredError()
self.configs.cv_loss_function = loss_function.MeanSquaredError()
def __init__(self):
super(MethodConfigs, self).__init__()
pc = create_project_configs()
self.z_score = False
self.quiet = False
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = pc.cv_loss_function
self.loss_function = pc.loss_function
self.use_validation = False
self.metric = 'euclidean'
self.use_saved_cv_output = False
def set_data_set(self, name, target_labels, source_labels, is_regression):
assert is_regression
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/' + name
self.data_name = name
self.results_dir = name
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray(target_labels)
self.source_labels = np.asarray(source_labels)
def set_pollution(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
s = 'pollution-[3 4]-500-norm'
#s = 'pollution-[60 71]-500-norm'
self.data_dir = 'data_sets/' + s
self.data_name = s
self.results_dir = s
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([0])
self.source_labels = np.asarray([1, 2, 3])
self.target_domain_order = np.asarray([1, 0])
self.source_domain_order = np.asarray([3, 2])
def set_boston_housing_transfer(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
if self.use_1d_data:
self.data_dir = 'data_sets/boston_housing(transfer)'
self.data_name = 'boston_housing'
self.results_dir = 'boston_housing'
else:
self.data_dir = 'data_sets/boston_housing-13(transfer)'
self.data_name = 'boston_housing-13'
self.results_dir = 'boston_housing-13'
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([0])
self.source_labels = np.asarray([1])
def set_concrete_transfer(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
if self.use_1d_data:
self.data_dir = 'data_sets/concrete-feat=0'
self.data_name = 'concrete-feat=0'
self.results_dir = 'concrete-feat=0'
else:
self.data_dir = 'data_sets/concrete-7'
self.data_name = 'concrete-7'
self.results_dir = 'concrete-7'
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([1])
self.source_labels = np.asarray([3])
def set_wine(self):
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
if self.use_1d_data:
self.data_dir = 'data_sets/wine-small-feat=1'
self.data_name = 'wine-small-feat=1'
self.results_dir = 'wine-small-feat=1'
else:
self.data_dir = 'data_sets/wine-small-11'
self.data_name = 'wine-small-11'
self.results_dir = 'wine-small-11'
self.data_set_file_name = 'split_data.pkl'
self.target_labels = np.asarray([0])
self.source_labels = np.asarray([1])
def set_synthetic_step_linear_transfer(self):
self.loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/synthetic_step_linear_transfer'
self.data_name = 'synthetic_step_linear_transfer'
self.data_set_file_name = 'split_data.pkl'
self.results_dir = 'synthetic_step_linear_transfer'
self.target_labels = np.asarray([0])
self.source_labels = np.asarray([1])
def set_data_set_defaults(self,
data_set_name,
target_labels=None,
source_labels=None,
is_regression=True):
assert is_regression
self.loss_function = loss_function.MeanSquaredError()
self.cv_loss_function = loss_function.MeanSquaredError()
self.data_dir = 'data_sets/' + data_set_name
self.data_name = data_set_name
self.results_dir = data_set_name
self.data_set_file_name = 'split_data.pkl'
self.target_labels = target_labels
self.source_labels = source_labels
if target_labels is not None:
self.target_labels = np.asarray(target_labels)
if source_labels is not None:
self.source_labels = np.asarray(source_labels)
def __init__(self, configs=base_configs.MethodConfigs()):
super(ScipyOptNonparametricHypothesisTransfer, self).__init__(configs)
self.cv_params['C'] = 10**np.asarray(range(-4, 4), dtype='float64')
self.g_nw = method.NadarayaWatsonMethod(configs)
self.g_nw.configs.target_labels = None
self.g_nw.configs.source_labels = None
self.g_nw.configs.cv_loss_function = loss_function.MeanSquaredError()
self.g_nw.quiet = True
self.k = 3
self.metric = configs.metric
self.bias = 0
self.use_huber = use_huber
def __init__(self, data_set=None, **kwargs):
super(VisualizationConfigs, self).__init__(data_set, **kwargs)
self.max_rows = max_rows
pc = ProjectConfigs(data_set)
self.copy_fields(pc, pc_fields_to_copy)
self.data_set_to_use = pc.data_set
self.title = bc.data_name_dict.get(self.data_set_to_use,
'Unknown Data Set')
self.show_legend_on_all = True
self.x_axis_string = 'Number of labeled instances'
self.ylims = None
self.generate_file_names(pc)
viz_loss_function = loss_function.MeanSquaredError()
self.always_show_y_label = True
is_regression = not self.data_set_to_use in classification_data_sets
instance_subset = 'is_train'
if not hasattr(self, 'loss_to_use'):
self.loss_to_use = loss_to_use
if self.loss_to_use == LOSS_Y:
if pc.use_var:
results_features = ['y', 'true_y']
self.y_axis_string = 'Variance Error'
else:
results_features = ['y', 'true_y']
self.y_axis_string = 'Prediction Error'
elif self.loss_to_use == LOSS_P:
results_features = ['p', 'true_p']
self.y_axis_string = 'P(X) Error'
elif self.loss_to_use == LOSS_NOISY:
results_features = ['is_noisy', 'is_selected']
viz_loss_function = loss_function.LossAnyOverlap()
self.y_axis_string = 'Noisy Error'
elif self.loss_to_use == LOSS_ENTROPY:
instance_subset = 'is_selected'
results_features = ['y_orig', 'y_orig']
viz_loss_function = loss_function.LossSelectedEntropy(
is_regression=is_regression)
self.y_axis_string = 'Selection Distribution Error'
else:
assert False
self.instance_subset = instance_subset
self.results_features = results_features
self.loss_function = viz_loss_function
value_housing = pricing_data[I, 1]
value_housing /= value_housing.max()
if apply_log:
value_housing = np.log(value_housing)
data = combine_data(loc_traffic, value_traffic, loc_housing, value_housing)
else:
I &= np.isfinite(locations[:, 0])
data = create_transfer_data(locations, pricing_data, I, apply_log)
print 'n: ' + str(I.sum())
# pricing_data[:] = 1
if run_state_tests:
m = base_configs.MethodConfigs()
m.cv_loss_function = loss_function.MeanSquaredError()
m.loss_function = loss_function.MeanSquaredError()
loss = loss_function.MeanSquaredError()
m.use_validation = True
m.target_labels = np.asarray([1])
m.source_labels = np.asarray([0])
stacking_transfer = transfer_methods.StackingTransfer(deepcopy((m)))
m.just_target = True
target_learner = far_transfer_methods.GraphTransfer(deepcopy(m))
m.just_target = False
m.just_transfer = True
source_learner = far_transfer_methods.GraphTransfer(deepcopy(m))
num_splits = 10
errors = np.zeros((all_states.size, 3))
for state_idx, s in enumerate(all_states):
